English

[!IMPORTANT] This is not affiliated with, officially maintained by, or endorsed by SenseAudio.

[!TIP] API keys available from the author. The author sells SenseAudio API keys. Note that the SenseAudio platform does not support prompt cache hits, so your quota is consumed at full input-token cost on every request — it doesn't last as long as cache-enabled platforms.

Integrate SenseAudio models into GitHub Copilot Chat as a VS Code extension.

Usage

1. Set API Key: Ctrl+Shift+P → SenseAudio: Manage API Keys
2. Show Models: Click the settings icon in the model picker → Language Models panel → set your desired models to Visible
3. Select Model: In the Copilot Chat bottom model picker, choose a "SenseAudio" model
4. Start chatting

Status Bar: Plan Usage & Token Indicator

Once installed, the status bar shows your SenseAudio plan usage for the current 5-hour window (e.g. 5H 65%, or -- before the first fetch). Hover the status bar item to see:

• Plan usage: 5-hour / weekly / monthly utilization (with absolute credits, e.g. 5H——65% (6,500 / 10,000 积分)), the 5-hour reset countdown, and the current billing mode.
• Balance: cash balance plus gift vouchers (with the earliest expiry date).
• Token usage: cumulative input/output token counts, plus the cumulative cache hit count and cache hit rate for models that return cache metrics in an OpenAI-compatible format.

When the monthly plan quota is exhausted, the main text switches to the balance (e.g. 余额 ¥358.78) so you can tell at a glance that you are now being billed from your balance.

Click the status bar item to refresh immediately.

[!IMPORTANT] Two separate billing rules (see the official docs):

• Period quotas (5h / weekly) are rate-limit windows. When exhausted, they simply wait for the next period to reset — they do not consume your balance.
• Plan credits (monthly) are the subscription quota. When exhausted, the overage policy applies: if balance auto-pay is enabled, the overage is billed per-use (vouchers → cash); otherwise the account downgrades to the Free plan.

The status bar only appears while you are actually using a SenseAudio model: it stays hidden on startup and when other chat model providers are in use, and auto-hides after 60 seconds of inactivity.

Relevant settings: senseaudio.showUsageInStatusBar (default true), senseaudio.showUsageInTooltip (default true), senseaudio.usageRefreshInterval (default 5 minutes), and senseaudio.enableThirdPartyTokenIndicator (default false — controls only the advanced token counter; the plan-usage display is independent).

[!NOTE] Whether non-DeepSeek models display cache data depends on whether the model API returns cache metrics in an OpenAI-compatible format. This does not indicate whether the model supports caching — caching support depends on SenseAudio.

Git Commit Messages

Click the magic wand button in the Source Control (SCM) panel to auto-generate a commit message.

You can configure the model, language, number of recent commits to reference, and whether to attach context files.

Model Temperature Presets

Switch temperature presets via the senseaudio.modelPreset setting (or set senseaudio.temperature / senseaudio.top_p directly with senseaudio.modelPreset set to "custom").

Built-in presets:

You can also configure senseaudio.temperature and senseaudio.top_p directly in settings.json (requires senseaudio.modelPreset set to "custom").

Extended Vision Understanding

This extension adds extended vision understanding capability to text-only models that do not natively support vision. When you send a message with an image to these models, they can call a vision-capable model to describe the image, and then answer based on that description.

You can configure the default vision model and whether to enable thinking when describing images. By default, qwen3.6-35b-a3b is used to describe images. Set senseaudio.visionProxyModel to any vision-capable model ID (e.g. qwen3.6-35b-a3b, qwen3.8-27b, deepseek-v4.1-flash, glm-5.3-flash).

Scope note — how images reach the vision proxy: the ask_image proxy applies to images you paste/attach manually into the chat (the extension declares imageInput: true so VS Code forwards image data to it, and non-vision models delegate to the vision proxy model). It does not apply to screenshots taken by VS Code's built-in screenshot tool (e.g. in agent mode) — screenshot analysis is handled internally by the Copilot Chat framework using GitHub Copilot's own vision models, which is outside a third-party provider's control. If your Copilot plan's vision model is unavailable, the built-in screenshot tool reports "vision model query unavailable"; this does not affect manual image pasting, which still works through the extension's proxy.

Model List

The extension ships with built-in definitions for the following SenseAudio chat models (verified against the live /v1/models endpoint on 2026-09-29; specs from the model page):

¹ The official docs table lists — for every spec of this model, and it is absent from models.dev / OpenRouter. 256K / 64K is assumed (same class as S2-Flash/Lite); revisit if the platform publishes real values.

All models support the OpenAI-compatible protocol. Protocol capability (Responses / Anthropic) is detected dynamically at startup from GET /v1/models — no model IDs are hardcoded. In auto mode, priority: Responses (if enableResponsesApi enabled) > Anthropic (if enableAnthropicApi enabled) > OpenAI. [!WARNING] The Anthropic protocol has compatibility issues with some models (e.g. DeepSeek: forced thinking + temperature/top_p returns 400 "请求参数组合无效"). The OpenAI-compatible format is recommended; use Anthropic only when you specifically need the native Messages format.

[!TIP] Automatic model discovery is enabled by default: the extension fetches the live model list from GET /v1/models and hides models that are not available on your account. Image-generation models (senseaudio-image-2.0, doubao-seedream-5-0, sensenova-u1-fast) are excluded from the picker.

Configuration

Available in settings.json:

{
  "senseaudio.apiMode": "auto",
  "senseaudio.commitLanguage": "auto",
  "senseaudio.commitModel": "glm-5.3-flash",
  "senseaudio.commitMessagePrompt": "",
  "senseaudio.requestTimeout": 600000,
  "senseaudio.recentCommitsCount": 10,
  "senseaudio.commitIncludeCommitDiff": false,
  "senseaudio.commitAttachContextFiles": true,
  "senseaudio.enableAutoModelDiscovery": true,
  "senseaudio.syncModelsOnStartup": true,
  "senseaudio.maxInputTokensRatio": 1.0,
  "senseaudio.enableThirdPartyTokenIndicator": true,
  "senseaudio.enableResponsesApi": false,
  "senseaudio.enableAnthropicApi": false
}

[!NOTE] Models with switchable thinking (e.g., DeepSeek, Qwen) provide reasoning effort levels such as Disabled/High/Maximum.

Build

npm install
npm run compile
npm run build      # packages extension.vsix

License

AGPL-3.0 License. This project builds upon the architecture of opencode-go-copilot (MIT) and oai-compatible-copilot (MIT).

中文

[!IMPORTANT] 本插件与 SenseAudio 无关，也未获得其官方维护或认可。

[!TIP] 作者有售 API Key。 作者出售 SenseAudio API Key。注意：SenseAudio 平台不支持缓存命中（prompt cache），每次请求都按完整输入 Token 计费，额度不如支持缓存的平台耐用。

将 SenseAudio 模型集成到 GitHub Copilot Chat 的 VS Code 插件。

使用

1. 设置 API Key：Ctrl+Shift+P → SenseAudio: 管理 API Keys
2. 显示模型：在模型选择器中点击设置图标 → 语言模型 面板 → 将需要使用的模型显示
3. 选择模型：在 Copilot Chat 底部模型选择器中选择 "SenseAudio" 下的模型
4. 开始对话

状态栏：套餐用量与 Token 指示器

安装后，使用 SenseAudio 提供的模型时，状态栏会显示套餐当前 5 小时窗口的用量（如 5H 65%，首次获取前显示 --）。悬停状态栏条目可查看：

• 套餐用量：5 小时 / 周 / 月三个窗口的使用率（含绝对积分，如 5H——65% (6,500 / 10,000 积分)）、5 小时窗口的重置倒计时，以及当前计费模式。
• 余额：现金余额 + 代金券（含最早到期日）。
• Token 用量：累计输入/输出 Token 量；当模型接口以 OpenAI 兼容格式返回缓存数据时，还会显示累计缓存命中量与缓存命中率。

当月度套餐额度耗尽时，主文本会切换为余额（如 余额 ¥358.78），一眼即可看出已开始从余额扣费。

点击状态栏条目可立即刷新。

[!IMPORTANT] 两套计费规则相互独立（详见官方文档）：

• 周期额度（5 小时 / 周）是限流窗口。耗尽后只需等待下一周期自动恢复，不会消耗余额。
• 套餐积分（月度）才是订阅额度。耗尽后走超额策略：若已开启余额自动支付，超出部分按量计费（代金券 → 现金余额）；否则账号降级至 Free 版。

状态栏仅在您实际使用 SenseAudio 模型时显示：启动时隐藏、使用其他模型提供商的模型时不显示，停止使用（空闲 60 秒）后自动隐藏。

相关设置：senseaudio.showUsageInStatusBar（默认 true）、senseaudio.showUsageInTooltip（默认 true）、senseaudio.usageRefreshInterval（默认 5 分钟）、senseaudio.enableThirdPartyTokenIndicator（默认 false，仅控制高级 Token 计数器，与套餐用量显示相互独立）。

[!NOTE] 非 DeepSeek 的模型是否显示缓存数据取决于模型接口是否通过 OpenAI 格式返回缓存数据，这并不代表此模型是否支持缓存。模型对于缓存的支持情况取决于 SenseAudio。

Git 提交消息

在源代码管理（SCM）面板中点击魔法棒按钮，自动生成 Git 提交消息。

可在配置里配置使用的模型、语言、参考的最近提交数量以及是否附加上下文文件。

扩展视觉理解

本插件为不支持视觉理解的纯文本模型添加了扩展视觉理解功能，当你向这些模型发送带有图片的信息时，他们可以调用支持视觉理解的模型为图片输出描述，然后再回答。

通过配置文件可更改默认使用的模型以及是否在描述图片时启用思考。默认情况下，将使用 qwen3.6-35b-a3b 描述图片。将 senseaudio.visionProxyModel 设为任意支持视觉的模型 ID（如 qwen3.6-35b-a3b、qwen3.8-27b、deepseek-v4.1-flash、glm-5.3-flash）即可。

适用范围说明 —— 图片如何到达视觉代理：ask_image 代理作用于你在聊天中手动粘贴/附带的图片（扩展声明 imageInput: true，VS Code 会将图片数据传给扩展，非视觉模型再委托给视觉代理模型）。它不覆盖 VS Code 内置截图工具拍摄的截图（如 agent 模式下的截图）——截图分析由 Copilot Chat 框架内部使用 GitHub Copilot 自带的视觉模型完成，第三方提供商无法接管。如果你的 Copilot 套餐的视觉模型不可用，内置截图工具会提示"视觉模型查询暂不可用"；这不影响手动粘贴图片，后者仍会通过扩展的代理正常工作。

模型列表

扩展内置了以下 SenseAudio Chat 模型定义（2026-09-29 对照实时 /v1/models 核实；规格取自模型页）：

¹ 官方文档对该模型的所有规格列均为 —，且 models.dev / OpenRouter 均未收录；暂按同类模型（S2-Flash/Lite）假定 256K / 64K，待平台公布真实值后修正。

所有模型均支持 OpenAI 兼容协议。协议能力（Responses / Anthropic）在启动时从 GET /v1/models 动态探测——不硬编码模型 ID。auto 模式下优先级：Responses（若开启 enableResponsesApi）> Anthropic（若开启 enableAnthropicApi）> OpenAI。 [!WARNING] Anthropic 协议对部分模型存在兼容性问题（如 DeepSeek：强制思考 + temperature/top_p 返回 400"请求参数组合无效"）。建议优先使用 OpenAI 兼容格式；仅在明确需要 Anthropic 原生 Messages 格式时使用。

[!TIP] 自动模型发现默认开启：扩展会从 GET /v1/models 拉取实时模型列表，隐藏你账号下不可用的模型。图片生成模型（senseaudio-image-2.0、doubao-seedream-5-0、sensenova-u1-fast）不会出现在选择器中。

调整模型温度

通过 senseaudio.modelPreset 设置切换温度预设（也可将 senseaudio.modelPreset 设为 "custom" 后直接配置 senseaudio.temperature / senseaudio.top_p）。

内置 4 个预设档位：

也可在 settings.json 中直接配置 senseaudio.temperature 和 senseaudio.top_p（需将 senseaudio.modelPreset 设为 "custom"）。

配置

可在 settings.json 中配置：

{
  "senseaudio.apiMode": "auto",
  "senseaudio.commitLanguage": "auto",
  "senseaudio.commitModel": "glm-5.3-flash",
  "senseaudio.commitMessagePrompt": "",
  "senseaudio.requestTimeout": 600000,
  "senseaudio.recentCommitsCount": 10,
  "senseaudio.commitIncludeCommitDiff": false,
  "senseaudio.commitAttachContextFiles": true,
  "senseaudio.enableAutoModelDiscovery": true,
  "senseaudio.syncModelsOnStartup": true,
  "senseaudio.maxInputTokensRatio": 1.0,
  "senseaudio.enableThirdPartyTokenIndicator": false,
  "senseaudio.showUsageInStatusBar": true,
  "senseaudio.showUsageInTooltip": true,
  "senseaudio.usageRefreshInterval": 5,
  "senseaudio.enableResponsesApi": false,
  "senseaudio.enableAnthropicApi": false
}

[!NOTE] 支持切换思考模式的模型（如 DeepSeek、Qwen）提供禁用思考/高/极高等推理强度选项。

编译

npm install
npm run compile
npm run build      # 打包为 extension.vsix

许可

AGPL-3.0 许可。本项目基于 opencode-go-copilot（MIT）与 oai-compatible-copilot（MIT）的架构实现。